AST + AI = Developer Co-Pilot 2.0
summary
Combining ASTs with LLMs produces concrete, actionable fixes instead of generic advice: code is parsed and navigated to exact functions, transformations are applied in the tree and emitted back as edits, and automated PRs can be generated and tested. Empirical tests across about 800 components show big cost and precision wins — one example yields 260 tokens vs ~1,500 tokens — and tooling uses Babel parser plus js-codemods and simple cloud API scripts. A working pipeline for AST-driven PRs exists and reduces false positives while locating exact lines to change.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:50AI struggles with raw code
- 4:34AST bridges AI understanding
- 7:05AST vs AI token efficiency
- 10:17API migrations with AST
- 12:01AST in code reviews
- 13:23Tooling and steps
- 14:29Key takeaways and future
- 15:33Q&A
related talks
21mAutomating Front-End Refactoring
26mOrchestrating AI Agents for Large-Scale Refactoring
34mMass Frontend Refactors: Less Stress; More Success
21mand AI
27mWill AI (Finally) Replace Developers?
46mCompilers, User Interfaces & the Rest of Us
01h 01m"Looks Good to Me": A Practical Guide to Handling AI-Generated Code
25mUtilizing the JavaScript Abstract Syntax Tree (AST) for Automated Frontend Lifecycle Management